LexAI - Production Legal Engine

LexAI is a high-performance, hybrid C++/Python system designed for automated legal adjudication under Croatian Law. It combines a deterministic C++ core for legal logic with a Python FastAPI orchestration layer, integrated with a PostgreSQL database and a local LLM (Ollama) for natural language processing.

Architecture

The system runs as a multi-container Docker application:

1. LexAI App: FastAPI + C++ Core (Business Logic).
2. PostgreSQL: Persistent storage for driver history (Recidivism checks).
3. Ollama: Local LLM service running mistral for parsing case descriptions.

Features

• C++ Core:
  • Speeding: Calculates fines/points based on config/rules.json. Checks DB for prior offenses (Recidivism doubles the fine).
  • Alcohol: Adjudicates based on driver category (Young, Professional, Standard).
  • Interest: Calculates penalty interest (Civil Obligations Act).
  • Severance: Calculates severance pay (Labor Act).
• AI Parsing: Uses a local LLM (mistral) to parse unstructured text (e.g., "Driver was drunk, 1.5 g/kg") into structured JSON.
• PDF Generation: Generates official verdict documents using Jinja2 and WeasyPrint.
• Configuration: All legal parameters (fines, limits, rates) are hot-loadable from config/rules.json.

Quick Start (Docker Compose)

This is the recommended way to run the full stack.

1. Start the System:

   docker compose up --build

   Note: The first run will take a few minutes to download the 4GB LLM model.

2. Access the UI: Open http://localhost:8000 in your browser.

3. Test the API:

   # Speeding (checks DB for recidivism)
   curl -X POST "http://localhost:8000/judge/speeding" \
        -H "Content-Type: application/json" \
        -d '{"measured_speed": 130, "limit": 100, "tolerance": 10, "driver_id": "recidivist_1"}'
   
   # AI Parsing
   curl -X POST "http://localhost:8000/parse" \
        -H "Content-Type: application/json" \
        -d '{"text": "Driver detected with 1.5 g/kg alcohol, young driver."}'

Directory Structure

• src/core/: C++ business logic (Speeding, Alcohol, Interest, Severance).
• src/api/: Python FastAPI app, PDF generator, LLM parser.
• config/: External configuration (rules.json).
• docker/: Dockerfile and entrypoint scripts.
• tests/: Unit tests.

Local Development

To develop locally, you need cmake, g++, python3.11, libpqxx-dev, and nlohmann-json3-dev.

1. Build C++ Core:

   mkdir build && cd build
   cmake .. && make

2. Run API:

   export PYTHONPATH=$PYTHONPATH:$(pwd)/build
   uvicorn src.api.main:app --reload

   Note: Local run requires a running Postgres instance and Ollama service accessible at the URLs defined in the code.